testArrayNumpyLabelled fails on Python 2.7.10

[cel4]

FAIL: testArrayNumpyLabelled (pandas.io.tests.test_json.test_ujson.NumpyJSONTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/ch/repo/pandas/.tox/py27/lib/python2.7/site-packages/pandas/io/tests/test_json/test_ujson.py", line 1141, in testArrayNumpyLabelled
    self.assertTrue((expectedvals == output[0]).all())
AssertionError: False is not true
-------------------- >> begin captured stdout << ---------------------
[[42 31]
 [24 99]
 [ 2 78]]
[[31 42]
 [99 24]
 [78  2]]

--------------------- >> end captured stdout << ----------------------

Order of dict elements seems to be non-deterministic for python2.7 versions as well.

[jreback]

well this doesn't fail on travis AFAICT, nor have I ever actually seen this fail. So not sure how to repro.

[kawochen]

Do you have a 32-bit build? This test does seem to depend on a few things, including architecture and the PYTHONHASHSEED environment variable.

[cel4]

@kawochen, No, I am running a 64bit interpreter. Yes, the failure is highly hardware/interpreter/seed dependent. But that is the point of this bug report.

[jreback]

@cel4 can you give the specific system this fails on. Yes its is somewhat non-deterministic, though in py2 the ordering is deterministic. Kind of a silly test actually (but IIRC came from another package :). Why don't you do a pull-request to sort them first.

[cel4]

INSTALLED VERSIONS
------------------
commit: None
python: 2.7.10.final.0
python-bits: 64
OS: Darwin
OS-release: 14.5.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: de_DE.UTF-8
LANG: de_DE.UTF-8

pandas: 0.16.2+340.ge4368de.dirty
nose: 1.3.7
Cython: 0.23
numpy: 1.9.2
scipy: None
statsmodels: None
IPython: None
sphinx: None
patsy: None
dateutil: 2.4.2
pytz: 2015.4
bottleneck: None
tables: None
numexpr: None
matplotlib: None
openpyxl: 1.8.6
xlrd: 0.9.4
xlwt: None
xlsxwriter: 0.7.3
lxml: 3.4.4
bs4: 4.4.0
html5lib: 0.999999
httplib2: 0.9.1
apiclient: 1.2
sqlalchemy: 1.0.8
pymysql: None
psycopg2: None

If the test fails or passes depends on the python hashseed.

e.g. PYTHONHASHSEED='1145934480' deterministically fails that test, PYTHONHASHSEED='2585294338' deterministically passes the test here.